Namespace import crashes the plugin

[InvictusMB]

• babel-plugin-macros version: 2.5.1

Relevant code or config

import * as foo from 'foo.macro'

What happened:
Compiler fails with the following message

./src/App.jsx
TypeError: Cannot read property 'name' of undefined
    at Array.map (<anonymous>)

Problem description:
ImportNamespaceSpecifier node is not handled properly here and the error message is not helpful. Offending code:

babel-plugin-macros/src/index.js

Lines 78 to 80 in e0ebca7

	s.type === 'ImportDefaultSpecifier'
	? 'default'
	: s.imported.name,

Suggested solution:
ImportNamespaceSpecifier should either be handled as default case or as a separate namespace case. Or maybe both variants are valid and there should be a config option akin to allowSyntheticDefaultImports in TypeScript

Quick fix:

(s.type === 'ImportDefaultSpecifier' || s.type === 'ImportNamespaceSpecifier')
  ? 'default'
  : s.imported.name,

[kentcdodds]

Thanks for bringing this up @InvictusMB. My inclination is to just throw a hard error with a sensible error message to tell people they can't use macros this way. But then I think why can't they? So instead, do you think that you could make a PR to make this work? I don't want to treat it as a default import though. I want it to work as it normally would (where the namespace specifier is basically an object with all exports on it). What do you think?

[InvictusMB]

I don't have a good answer to this.

I want it to work as it normally would (where the namespace specifier is basically an object with all exports on it).

I agree, however I don't see much syntactic difference between default import and namespace import for a macro user.
The only significant thing is that it is illegal to use an imported namespace as a function.
Other than that they are functionally equivalent. Especially so, when the majority of npm modules are still bundled as CommonJS modules and in es6 you can import them either way.

My gut feeling tells me that it is better to treat default imports as equal to namespace imports and not complicate 99% of use cases.

Otherwise it seems to be a rabbit hole. Let's say we have this code

import * as foo from 'foo.macro';

foo.bar();
foo["baz"]();

I assume, you want it to be handled as equivalent to

import {bar, baz} from 'foo.macro';

bar();
baz();

The algorithm could be:

• When traversing the program, separate namespace specifiers from the other cases
• When applying the macro
  • follow each namespace reference
  • if reference is a child of MemberExpression or JSXMemberExpression
    • follow property path
    • if property is Identifier save name as a name
    • if property is Literal save value as a name
  • if reference is Identifier or JSXIdentifier then fallback to default name or throw (???)
  • pass calculated name to the macro
  • either pass parent path to the macro
  • or pass reference path as is and let the macro figure out where it points to.

On the macro side this will also complicate things.
With named import bar will always be an Identifier or JSXIdentifier.
With namespaced import we will get a reference named bar but the path can either be MemberExpression or JSXMemberExpression if we passed parent path.
Otherwise it can be Identifier or JSXIdentifier and the macro has to check if it is under MemberExpression or JSXMemberExpression and use parent path instead of provided path for replacement.
I'm afraid this will be confusing both ways.

I'm not 100% sure this extra complexity is worth the outcome.

P.S. The test suite fails on my windows machine. Does it have any windows specific reasons to fail?

λ npm test

> [email protected] test C:\Projects\babel-plugin-macros
> kcd-scripts test
 PASS  src/__tests__/create-macros.js
 FAIL  src/__tests__/index.js
  ● Test suite failed to run

    Cannot find module 'babel-plugin-macros-test-fake/macro' from 'C:\Projects\babel-plugin-macros\src\__tests__'

      at Function.module.exports [as sync] (node_modules/resolve/lib/sync.js:69:15)
      at nodeResolvePath (node_modules/babel-plugin-macros/dist/index.js:49:18)
      at applyMacros (node_modules/babel-plugin-macros/dist/index.js:178:23)
      at VariableDeclaration.path.get.filter.forEach.child (node_modules/babel-plugin-macros/dist/index.js:126:30)
          at Array.forEach (<anonymous>)
      at VariableDeclaration (node_modules/babel-plugin-macros/dist/index.js:116:55)
      at NodePath._call (node_modules/@babel/traverse/lib/path/context.js:53:20)
      at NodePath.call (node_modules/@babel/traverse/lib/path/context.js:40:17)
      at NodePath.visit (node_modules/@babel/traverse/lib/path/context.js:88:12)
      at TraversalContext.visitQueue (node_modules/@babel/traverse/lib/context.js:118:16)

[kentcdodds]

I'm also not convinced it's worth the work.

Tests should definitely work on windows, but I don't have a windows machine to test it on. PRs welcome for that!

[InvictusMB]

I have a PR on the way for this.
So the suggested plan is:

1. For now namespace imports will be treated the same way as default imports. There is no reason not to as the only semantic difference is that a namespace is not callable. But it is not the job of this plugin to enforce that. TypeScript and linters can handle the semantics for us.

2. If the demand for a separate handling of namespaces appears it can be later introduced together with a config switch. I believe in this case the plugin would have to export SymbolNamespace symbol for macros to identify namespace imports as we cannot simply use namespace since it is not a reserved word.
   SymbolNamespace could be used here
   

   babel-plugin-macros/src/index.js

   Lines 97 to 100 in 2d57c60

   	importedName:
   	s.type === 'ImportDefaultSpecifier'
   	? 'default'
   	: s.imported.name,

3. If there is a need to track foo.bar usages to make them equivalent to import {bar} then a separate helper can be released. It will probably require some additional config from macro creators for handling of corner cases. But for the scope of this plugin it is a bit too much of complexity. The complexity comes from the fact that imported bar is a standalone reference but namespaced foo.bar is an expression and needs a different treatment inside the macro itself. One way to workaround that could be to have 2 runs where the first will transform import * as foo into import {bar as <uniqueId>} based on references further in the program and then the second run will do the regular handling of import specifiers.

[kentcdodds]

You know what, I'd really rather wait until someone's really asking for this before making a bunch of changes 😬

[InvictusMB]

I agree. I want the plugin not crashing though.
And I believe we have enough arguments to fix this in the least effort way.

[InvictusMB]

@kentcdodds
Was this fixed?

[kentcdodds]

No, but nobody's actively working on it or asking for it. Sorry, I should have commented about that when closing.

[kentcdodds]

You know what, I'll go ahead and work on this myself.